Davis SR, Davison SL, Donath S, Bell RJ. Circulating Androgen Levels and Self-reported Sexual Function in Women. JAMA. 2005;294(1):91-96. doi:10.1001/jama.294.1.91
Author Affiliations: Women’s Health Program,
Department of Medicine, Monash Medical School, Alfred Hospital, Victoria,
Australia (Drs Davis and Bell and Ms Donath); Jean Hailes Foundation, Victoria,
Australia (Drs Davis, Davison, and Bell); Department of Biochemistry, Monash
University, Victoria, Australia (Dr Davison); and Murdoch Children’s
Research Institute and Department of Paediatrics, University of Melbourne,
Melbourne, Australia (Ms Donath).
Context It has been proposed that low sexual desire and sexual dysfunction are
associated with low blood testosterone levels in women. However, evidence
to support this is lacking.
Objective To determine whether women with low self-reported sexual desire and
sexual satisfaction are more likely to have low serum androgen levels than
women without self-reported low sexual desire and sexual satisfaction.
Design, Setting, and Participants A community-based, cross-sectional study of 1423 women aged 18 to 75
years, who were randomly recruited via the electoral roll in Victoria, Australia,
from April 2002 to August 2003. Women were excluded from the analysis if they
took psychiatric medication, had abnormal thyroid function, documented polycystic
ovarian syndrome, or were younger than 45 years and using oral contraception.
Main Outcome Measures Domain scores of the Profile of Female Sexual Function (PFSF) and serum
levels of total and free testosterone, androstenedione, and dehydroepiandrosterone
Results A total of 1021 individuals were included in the final analysis. No
clinically significant relationships between having a low score for any PFSF
domain and having a low serum total or free testosterone or androstenedione
level was demonstrated. A low domain score for sexual responsiveness for women
aged 45 years or older was associated with higher odds of having a serum dehydroepiandrosterone
sulfate level below the 10th percentile for this age group (odds ratio [OR],
3.90; 95% confidence interval [CI], 1.54-9.81; P = .004).
For women aged 18 to 44 years, having a low domain score for sexual desire
(OR, 3.86; 95% CI, 1.27-11.67; P = .02),
sexual arousal (OR, 6.39; 95% CI, 2.30-17.73; P<.001),
and sexual responsiveness (OR, 6.59; 95% CI, 2.37-18.34; P<.001) was associated with having a dehydroepiandrosterone sulfate
level below the 10th percentile.
Conclusions No single androgen level is predictive of low female sexual function,
and the majority of women with low dehydroepiandrosterone sulfate levels did
not have low sexual function.
Sexual dysfunction, primarily low libido, is common among women, with
prevalences of 8% to 50% previously reported.1- 3 The
prevalence appears to increase with age from the third decade3 as
well as after oophorectomy.4 Although multiple
psychosocial and health factors contribute to low sexual desire and arousal,5 it has been proposed that endogenous androgen levels
are significant independent determinants of sexual behavior in women.6,7 Most studies support a therapeutic
benefit of testosterone for women experiencing hypoactive sexual desire disorder,
and there is increasing use of testosterone for this purpose.8- 10 It
is widely believed that a low serum free testosterone level is the diagnostic
marker for the cluster of symptoms described as characterizing “female
androgen insufficiency”6 based on therapeutic
trials8- 11 and
expert opinion.6,7,12 However,
evidence that a low serum testosterone level distinguishes women with low
sexual function from others, and that androgen deficiency syndrome can be
defined biochemically, is lacking. Therefore, we have investigated whether
low self-reported sexual function, assessed using the validated Profile of
Female Sexual Function (PFSF)13,14 in
women aged 18 to 75 years randomly recruited from the community, is associated
with low serum androgen levels.
Women were recruited by the random selection method using an electoral
roll database for Victoria, Australia. In Australia, where voting is compulsory,
every adult is registered on this roll. The city of Melbourne includes 26.25
electoral areas and rural Victoria includes 10.75 electoral areas. Each electoral
area was divided into sampling points of approximately equal numbers of 25 000
each. Melbourne had 105 sampling points and rural Victoria had 43 sampling
points. Starting addresses were selected at random from the electoral roll
for each of the sampling points. Interviews were conducted in person on Saturdays
and Sundays between 9:00 AM and 4:00 PM. Eight
interviews were conducted per sampling point and only 1 eligible person was
recruited per household.
Women were contacted by telephone. Women who were between the ages of
18 and 75 years were invited to participate in the study. Women were excluded
during telephone screening if they were pregnant or less than 6 weeks postpartum
or if they had experienced any of the following in the preceding 3 months:
an acute psychiatric illness; acute renal, liver, cardiovascular disease,
or any other acute major illness that would impair overall health and well-being;
gynecological surgery; active malignancy or cancer treatment, excluding nonmelanotic
We further excluded from this analysis women who had potentially confounding
conditions including current use of antidepressants, psychiatric medications,
or epilepsy medication, abnormal thyroid function (abnormal thyroid stimulating
hormone plus abnormal free thyroxine), polycystic ovarian syndrome, and use
of the oral contraceptive pill among women younger than 45 years. Polycystic
ovarian syndrome was identified on the basis of combinations of menstrual
history, ratio of luteinizing hormone to follicle-stimulating hormone higher
than 2, calculated free androgen index higher than 4.5, and sex hormone binding
globulin lower than 30 nmol/L.
Women provided fasting morning blood samples on the day they completed
and returned their questionnaires. Premenopausal women had blood drawn after
cycle day 8 and before menstruation to avoid the early follicular phase testosterone
nadir. We did not differentiate between mid-follicular and the luteal phase
days because the variation across these parts of the cycle in free and total
testosterone is minimal.15
This study was approved by the Southern Health Human Research and Ethics
Committee, Clayton, Australia. All participants provided written informed
The PFSF is a psychometrically validated instrument developed specifically
for the measurement of low sexual desire and related symptoms. It consists
of the 7 domains of desire, arousal, orgasm, pleasure, sexual concerns, responsiveness,
and self image and has no total score.13,14 We
selected the PFSF because it was developed based on input from women in the
community to ensure its relevancy and accuracy for symptoms, feelings, behaviors,
and attitudes. The validation of this questionnaire involved premenopausal
and postmenopausal Australian women, with specific attention to the linguistic
validity in Australian women.13
Fasting serum samples were stored at −80°C until assayed.
Total testosterone was measured by a highly sensitive direct manual radioimmunoassay
(Biosource Europe SA, Nivelles, Belgium) in the laboratory of Mayne Health
Dorevitch Pathology (Melbourne, Australia) using antibody-coated tubes and
500 μL of iodine-labeled T tracer. For 100 participants, between-batch
coefficients of variation were 12.8% at 4.89 ng/dL (0.17 nmol/L), 9.7% at
17.58 ng/dL (0.61 nmol/L), 8.8% at 51.01 ng/dL (1.77 nmol/L), and 7.1% at
331.41 ng/dL (11.5 nmol/L). For 20 participants, within-batch coefficients
of variation were 10.5%, 5.3%, 4.2%, and 4.7% at the same concentrations,
respectively. Samples with values below 5.76 ng/dL (0.2 nmol/L) (3.1% of all
samples) were reported as less than 5.76 ng/dL, but for statistical analysis
we assigned these samples a value of 2.88 ng/dL (0.1 nmol/L) because all calculations
were performed using Système International units. We reassessed the
performance of this assay against a validated and widely published radioimmunoassay
following organic solvent (ratio of solution of 3 parts ethylacetate to 2
parts hexane) extraction and celite column chromatography16- 18 and
were satisfied with a high level of consistency between assays. Free testosterone
was calculated using the Sodergard equation as previously described.19 Dehydroepiandrosterone sulfate (DHEAS) and sex hormone
binding globulin were measured using a solid-phase, 2-site chemiluminescent
enzyme immunometric assay with the Immulite 2000 automated analyzer (Diagnostic
Products Corporation, Los Angeles, Calif). The intra-assay and interassay
coefficients of variation for sex hormone binding globulin are 6.5% and 8.7%,
respectively; the detection limit is 0.2 nmol/L. For DHEAS level, the intra-assay
coefficient of variation is between 6.8% and 9.5% and the interassay coefficient
of variation is between 9.2% and 12.7%. Androstenedione was measured by direct
radioimmunoassay (DSL Inc, Webster, Tex). Follicle-stimulating hormone, thyroid-
stimulating hormone, luteinizing hormone, and prolactin were measured using
the Vitros ECI machine (Johnson & Johnson, Clinical Diagnostics Division,
Other community-based studies suggest a prevalence of self-reported
sexual dysfunction among women ranging from 8% to 50%.1- 3 Therefore,
we powered this study with a conservative estimate that 10% of the study population
would report low sexual function and that women with low sexual function would
be twice as likely to have a low androgen level (defined as less than the
10th percentile) than other women. Based on these assumptions our estimated
required sample size was 1100 (α = .05; 1−β = .80).
The distribution of the domain of sexual concerns differed from the other
domains with more than 20% of women in each age group reporting the maximum
score. Therefore, we excluded the domain of sexual concerns from our analysis.
The proportion of women reporting a zero score (from a possible range
of 0-100) for sexual satisfaction varied for the domains of the PFSF from
2.7% for responsiveness to 12.8% for sexual arousal.
The number of women who answered the questions varied slightly. The
PFSF domain scores were not normally distributed. Because the pattern of self-reported
sexual satisfaction differed according to age, data analyses were stratified
by age (<45 years vs ≥45 years). The decline in testosterone is greatest
between the third and fifth decades,20 does
not change during menopause,21 and changes
very little with age in postmenopausal women,22 providing
additional support for this approach.
For all the PFSF domains among older women, the proportion of women
reporting a score of zero (of a possible 100) determined the “low”
sexual satisfaction score (between 4% and 17% for each domain). For younger
women, less than 4.8% reported zero for 2 of the PFSF domains. Hence, to provide
consistency and avoid overdiagnosing low sexual function, we set a level of
5% for the category of low in the other 4 domains for younger women.
We looked at the relationships between the dichotomous variable (low
vs not low sexual function) and the series of continuous variables, which
were the androgen levels. Receiver operating characteristic (ROC) curves were
used to compare the relationship between sensitivity and specificity for different
cutoff levels for each of the androgens. Whether each androgen was useful
for discriminating between individuals with and without low sexual function
for each PFSF domain was established by testing whether the area under each
ROC curve was significantly different from 0.50.23
For each of the androgens, where the area under the ROC curve was highly
statistically significantly different from 0.50 (P≤.01),
the table of coordinate points for the ROC curve was used to identify a cutoff
point that would best differentiate those women with low sexual function from
those without low sexual function. The cutoff points were used to generate
contingency tables and to calculate odds ratios.
Of a total of 18 021 women, 15 621 were successfully contacted.
Of these, 8807 women declined to participate. Of those who wished to participate,
2853 were excluded because their age category was already full and 224 were
excluded based on the exclusion criteria. However, only 1423 of the 3737 who
had stated they wished to participate attended their study visit. Of the 1423
women recruited to the study, 199 were excluded (Figure). Of the remaining 1224 women, 1147 answered at least 1 PFSF
domain. Of these, 126 younger women were excluded on the basis of oral contraceptive
use, leaving a total of 1021 in the analysis. The characteristics of the study
population appear in Table 1.
The ROC curves provided no evidence for total or free testosterone being
useful for discriminating between individuals with or without low sexual function
for a PFSF domain for younger or older women (Table 2 and Table 3). In contrast,
the areas under the ROC curves for DHEAS levels were highly significantly
different from 0.50 (P≤.01) for older women in
relation to the domains of arousal, responsiveness, and pleasure, and for
younger women, for desire, arousal, and responsiveness. For older women, the
area under the ROC curve for androstenedione and pleasure was also highly
significantly different from 0.50.
For some of these highly significant associations in older women for
domains where 16.5% or 17.7% of women reported a zero score (DHEAS levels
and the domains of arousal and pleasure; androstenedione and pleasure), although
a serum androgen level cutoff was identified from the coordinate table of
the ROC curve, the cutoff resulted in at least 30% of women being classified
as having a low DHEAS level or androstenedione level.
In contrast, for the domain of sexual responsiveness in older women,
the reporting of a zero score by 4.1% was associated with a DHEAS level of
less than 294.8 ng/mL (0.8 μmol/L), which was below the 10th percentile
for this age group. The odds ratio for having a DHEAS level below 294.8 ng/mL
(0.8 μmol/L) in older women with low sexual responsiveness compared with
those without was 3.90 (95% confidence interval [CI], 1.54-9.81; Table 4).
For younger women, the best cutoff suggested by the ROC curve for each
of the domains desire, arousal, and responsiveness was a DHEAS level below
773.8 ng/mL (2.1 μmol/L), which corresponded to the 10th percentile for
this age group. The odds ratio for having a DHEAS level below 773.8 ng/mL
(2.1 μmol/L) in younger women and being below the fifth percentile for
the PFSF domains of desire were 3.86 (95% CI, 1.27-11.67); arousal, 6.39 (95%
CI, 2.30-17.73); and responsiveness, 6.59 (95% CI, 2.37-18.34) compared with
women who were above the fifth percentile for these domains (Table 4).
Multiple factors influence sexual behavior. The aim of this study was
to explore whether women with low self-reported sexual well-being are more
likely to have low serum androgen levels than women without self-reported
low sexual well-being.
The PFSF, although validated, had not previously been applied to a large
population of women so we were unable to anticipate the pattern of response.
The pattern of response determined the definition of low sexual function for
older women, that is, the reporting of a zero score, and provided a justifiable
cutoff of 5% for the category of low sexual function in younger women.
We found no evidence of associations between low scores for any of the
sexual domains evaluated and low serum total and free testosterone levels.
In contrast, we observed significant associations between low sexual desire,
arousal, and responsiveness in younger women and low responsiveness in older
women and low serum DHEAS level relative to age.
In light of the complexity of sexual function, it is not surprising
that the areas under the ROC curves that achieved statistical significance
for DHEAS level were not greatly different from 0.50. Exploration of cutoffs
identified from the ROC curves suggested that the likelihood of finding a
clinically useful association between women identified as having a low sexual
function and a low androgen level was greatest when the proportion of women
with low sexual function was small (<fifth percentile) and the normal range
for the serum androgen level was relatively large, such as the DHEAS level
among young women.22
The main strengths of this study are that we used a validated instrument
that demonstrates good discrimination between different levels of self-reported
sexual function in younger and older women13 to
evaluate sexual function, a sensitive radioimmunoassay for the measurement
of testosterone, and a community-based population of women from across our
state. A potential weakness is that despite our study population being recruited
by a random process based on the electoral roll, only 9.1% of those contacted
actually participated, indicating that selection pressures were operating,
as in any voluntary research project. Specific barriers to participation were
the requirements to provide a fasting blood sample, which for participants
from regional Victoria, constituting one third of our study population, involved
travel to a collection center, and for young women, timing of collection according
to their menstrual cycle. The approach we took was to minimize bias. The alternative
approach would have been to recruit a convenience sample, which may have achieved
a higher participation rate. However in doing so, we would have risked recruiting
a biased sample.
A concern might also be that a single serum testosterone level does
not reflect serum testosterone over time. Apart from the diurnal variation
in testosterone, there is no reason why serum testosterone should vary significantly
over days to weeks in postmenopausal women. For premenopausal women, small
variations across the mid-follicular and luteal phases would not have affected
classification with respect to the lowest 10th percentile. Furthermore, in
practice, clinical assessment will usually be made on a single serum sample.
Therefore, we believe a single early morning serum sample provided a practical
classification of women for the purpose of this study.
The hormone we identified as being associated with low self-reported
sexual function is DHEAS and not free testosterone. This is most likely due
to differing circulating levels of these steroids and the complexity of androgen
metabolism. DHEA is the most abundant sex steroid in women24 and
circulating DHEA and its sulfate, DHEAS, provide a large precursor reservoir
for the intracellular production of both estrogens and androgens.25- 27 Traditionally, circulating
hormone levels have been used as the main indicators of tissue exposure. However,
intracrinology plays a pivotal role in androgen metabolism, such that the
active androgens exert their effects in the same cells in which they are synthesized,
without release into the pericellular compartment.28 DHEA
and DHEAS are converted in extragonadal target tissues, such as the brain,
bone, and adipose, either to androstenedione or testosterone that may then
be aromatized to estrone or estradiol or converted by 5α-reductase to
dihydrotestosterone in the same cells.25,29 Thus
androgenic effects vary according to individual variations in the amount and
activity of the enzymes 5α-reductase and aromatase, and individual differences
in the androgen-receptor response. With substantial androgen production and
metabolism being intracrine, measurement of serum testosterone does not provide
a specific measure of androgen tissue exposure or action.
In addition to demonstrating that the measurement of testosterone is
not useful for the diagnosis of the proposed female androgen insufficiency
syndrome,6 our findings also do not support
a diagnostically useful role for the measurement of DHEAS. This is because
despite the increased likelihood that women with low sexual function have
a low DHEAS level, the majority of women with a low DHEAS level did not report
low sexual function.
Our results are not in conflict with testosterone being used pharmacologically
to treat hypoactive sexual desire disorder,9,10 nor
do they provide support for efficacy of DHEA therapy. Rather, our data, taken
together with what is already known about the intracrine physiology, suggest
that sex steroids influence female sexual function, but that there is no serum
androgen level that defines female androgen insufficiency. The measurement
of serum testosterone, free testosterone, or DHEAS in individuals presenting
with low sexual function is not informative and levels of these hormones should
not be used for the purpose of diagnosing androgen insufficiency in women.
Corresponding Author: Susan R. Davis, MD,
PhD, Women’s Health Program, Central and Eastern Clinical School, Monash
University, Alfred Hospital, Commercial Road, Prahran, Victoria, 3181 Australia
Author Contributions: Dr Bell had full access
to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Davis, Davison, Donath,
Acquisition of data: Davis, Davison.
Analysis and interpretation of data: Davis,
Davison, Donath, Bell.
Drafting of the manuscript: Davis, Donath,
Critical revision of the manuscript for important
intellectual content: Davis, Davison, Donath, Bell.
Statistical analysis: Donath, Bell.
Obtained funding: Davis.
Administrative, technical, or material support:
Study supervision: Davis.
Financial Disclosures: Dr Davis has received
unrestricted grant support from Procter & Gamble Pharmaceutical, Acrux,
and Solvay; and has acted as a consultant or has been on the advisory board
of Cellergy, Vivus, Acrux, and Organon. No other authors reported financial
Funding/Support: This research was funded by
grants 219279 and 284484 from the National Health and Medical Research Council
of Australia and by a philanthropic grant to the Jean Hailes Foundation from
Sue Ismiel and Daughters. The Profile of Female Sexual Function Questionnaire
was provided to the investigators by Procter & Gamble.
Role of the Sponsor: The funding organizations
and providers of the questionnaire were not involved in any aspect of the
design and conduct of the study; collection, management, analysis, and interpretation
of the data; and preparation, review, or approval of the manuscript.
Acknowledgment: We thank J. Montalto, PhD,
for his assistance with the biochemical analyses.